Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Context sensitivity in the spelling of English vowelsq
Rebecca Treiman,* Brett Kessler, and Suzanne Bick
Psychology Department, Wayne State University, 71 W. Warren Ave., Detroit, MI 48202, USA
Received 27 July 2001; revision received 26 September 2001
Abstract
Statistical analyses of English sound-to-spelling correspondences (Kessler & Treiman, 2001) show
that vowel spellings become more predictable, in some cases, when the preceding and following con-
sonants are taken into account. In four experiments, we asked whether adult spellers are sensitive to
such associations. We found evidence for sensitivity to associations involving both preceding and fol-
lowing consonants when examining adults� spellings of vowels in nonwords (Experiments 1 and 2) and
their substitution errors on vowels in real words (Experiment 3). The results show that phoneme-to-
grapheme mapping is sensitive to a broader array of context than just rime context. Additional findings
suggest that the context must be within the same syllable to be influential (Experiment 4). To the extent
that rimes play a special role in spelling, this role may derive from the fact that associations between
vowels and codas are more common in English than associations between vowels and onsets, not from
spellers� greater sensitivity to within-rime associations. � 2002 Elsevier Science (USA). All rights
reserved.
Keywords: Spelling; Sound-to-spelling correspondence; Vowels; Syllables; Rimes
To be literate, one must know how to read
words and how to spell them. There has been a
great deal of research on how adults derive the
pronunciations of written words in reading. This
topic is of particular interest because people who
are literate in an alphabetic writing system can
pronounce not only words they have seen before
but also words they encounter for the first time.
For instance, a reader who has never seen the
word glebe will probably pronounce it, correctly,
as /glib/. This ability shows that knowledge about
the spelling-to-sound relationships of the lan-
guage is represented, in some way, in the mind of
the reader. Researchers have studied the nature of
this knowledge by examining people�s pronuncia-
tions of nonwords (e.g., Andrews & Scarratt,
1998). They have also examined mispronuncia-
tions of real words to gain further insight into the
processes used in spelling-to-sound translation.
For example, the fact that great is occasionally
mispronounced as =g rit= in speeded pronuncia-
tion tasks has been interpreted to mean that
readers use their knowledge of the most common
pronunciation of ea, /i/ (e.g., Glushko, 1979).
Journal of Memory and Language 47 (2002) 448–468
www.academicpress.com
Journal ofMemory and
Language
qThis research was supported by NSF Grant SBR-
9807736. Helpful comments were provided by Michael
McCloskey, two anonymous reviewers, and audiences at
the University of Massachusetts and Haskins Labora-
tories.* Corresponding author. Present address: Psychology
Department, Washington University, Campus Box 1125,
One Brookings Drive, St. Louis, MO 63130-4899. Fax:
+314-935-7588.
E-mail address: [email protected] (R. Trei-
man).
0749-596X/02/$ - see front matter � 2002 Elsevier Science (USA). All rights reserved.
PII: S0749-596X(02 )00010-4
Our abilities in spelling are no less impressive
than our abilities in reading. A person who has
heard a piece of land referred to as a glebe but
who has never seen the word in print might spell it
as glebe, gleeb, or gleab. Only one of these options
is correct, but all three of them reveal the speller�sknowledge about the sound-to-letter mappings of
English. Although many studies have investigated
adults� knowledge and use of spelling-to-sound
relationships in English and other alphabetic
writing systems, relatively few studies have inves-
tigated sound-to-spelling translation. Likewise,
there is much less modeling work on spelling than
on reading. The goal of the present study was to
gain further information about adults� knowledge
of English sound-to-spelling relationships, infor-
mation that can help constrain models of the
spelling process and provide a foundation for
studies of how this knowledge is acquired.
Of particular interest, in this study, is the
spelling of vowels. In English, the spellings of
vowel phonemes are more variable than the
spellings of consonants (e.g., Kessler & Treiman,
2001). Likewise, adults make more substitution
errors on vowels than on consonants (e.g., Fi-
scher, Shankweiler, & Liberman, 1985). We ask
how knowledge of sound–spelling relationships is
represented for vowels and whether adults use
information about the neighboring consonants to
facilitate their spelling of vowels. Does conso-
nantal context influence the spelling of vowels? If
so, which aspects of the context are important?
Studies that have shown a special bond be-
tween the vowel and the following consonant(s) of
an English syllable provide a tentative answer to
the question of which aspects of context may in-
fluence vowel spelling. The link between a vowel
and the following consonant is often formalized
by saying that the vowel and the consonant (or
coda) belong to the rime constituent of the syllable
(e.g., Fudge, 1969). The initial consonant or
cluster is thought to belong to a separate con-
stituent of the syllable, the onset. Evidence for an
onset–rime division of the English syllable comes
from several sources. For example, English
speakers prefer to divide syllables at the onset–
rime boundary in metalinguistic tasks (e.g., Trei-
man, 1983), and these units play a role in memory
for speech (e.g., Treiman & Danis, 1988a) and
phonological organization (e.g., Kessler & Trei-
man, 1997). Also, people appear to use letter
groups that correspond to the rimes of spoken
syllables when translating from print to speech. A
number of studies have found evidence that these
units are influential in reading, together with evi-
dence that onset + vowel sequences are less influ-
ential (e.g., Taraban & McClelland, 1987;
Treiman, Mullennix, Bijeljac-Babic, & Richmond-
Welty, 1995; Treiman & Zukowski, 1988). Rime
units play a role in several theories of printed
word recognition (e.g., Zorzi, Houghton, & But-
terworth, 1998). Given the importance of rimes in
phonological processing and reading, one might
expect them to play a special role in spelling. In
this view, a given vowel may be spelled differently
when it occurs before certain codas than when it
occurs before other codas. The preceding conso-
nant should not have an effect, as it belongs to a
separate constituent of the syllable. According to
this rime constituency hypothesis, people weigh the
spelling options for a given phoneme according to
the rime context in which the phoneme appears.
Treiman and Zukowski (1988) found evidence
for the rime constituency hypothesis in a series of
experiments with college students. In one study,
participants were asked to spell nonwords such as
=tSend= and =f reh=. Of interest was how often
people used the ie spelling of =e=, which is found
only in the real word friend. Spellings with ie were
more common for nonwords such as =tSend=,which shares its rime with friend, than for non-
words such as =f reh=, which shares its onset and
vowel with friend. In another study, participants
rated chiend as a reasonably good spelling of
=tSend=, giving it higher marks than frieth as a
spelling of =f reh=. These results suggest that the
spelling of a medial vowel is affected by the
identity of the coda but not by the identity of
the onset and that rimes play a special role in the
spelling of English. However, the effects observed
in this study were relatively small.
Barry and Seymour (1988), unlike Treiman
and Zukowski (1988), found no evidence for a
special role of rimes in spelling production. Barry
and Seymour looked at nonwords such as /ten/,
for which the most common spelling of the vowel
in the nonword�s rime (ai) is not the most common
spelling of the vowel overall (a followed by final
e). They compared such items to nonwords such
as =dut=, for which the most common spelling of
the vowel in the particular rime (oo) is also the
most common spelling overall. In both cases,
participants tended to produce the vowel spelling
that was most common in the lexicon as a whole.
There was no significant difference between the
two types of nonwords in this regard, although
there was a trend in the direction of a rime effect.
Barry and Seymour (p. 26) concluded, on the basis
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 449
of these results, that ‘‘subjects . . . utilize sound-to-
spelling contingencies abstracted from the lexicon
as a whole rather than only from the subset of
rhyming words to generate their nonword spell-
ings.’’ That is, people store and use information
about the frequencies of various spelling options
for each phoneme; these frequencies are global
(reflecting how the phoneme is spelled in all rimes)
rather than local (reflecting how the phoneme is
spelled in particular rimes). However, the analysis
that led to this conclusion was post hoc and in-
volved a fairly small number of items.
Perry, Ziegler, and Coltheart (2002) seconded
the idea that spelling production involves pri-
marily phoneme units, not rime-based units.
These investigators designed a series of experi-
ments to investigate the effects of rime context in
spelling. Some of the nonwords in the experiments
were similar to those analyzed by Barry and
Seymour (1988) in that one of the phonemes in the
rime was generally spelled one way in that par-
ticular rime but a different way in the lexicon as a
whole. An example from the Perry et al. study is /
skof/, where /o/ is usually spelled as o followed by
final e but is more often spelled as oa in /of/. For
another type of nonword, also similar to a type
analyzed by Barry and Seymour, the most com-
mon spelling of the grapheme in the particular
rime was the most common spelling overall. An
example is /plem/, where a followed by final e is
the most common overall spelling of /e/ and also
its most common spelling in /em/. Perry et al. in-
cluded a third type of nonword for which the rime
had two spellings that were roughly equal in fre-
quency but one spelling of the critical phoneme
was more common overall than the other. With
=tSin=, for example, -ean and -een are roughly
similar in frequency but ea is the most common
spelling of /i/ in all rimes.
In one experiment, Perry et al. (2002) asked
participants to spell the nonwords using the first
reasonable spelling that came to mind. This
spelling production task is similar to that of Barry
and Seymour (1988). Participants in this experi-
ment generally used the most common overall
spelling for the phoneme. This tendency was only
weakly modified by the frequency of the pho-
neme�s spelling in the particular rime in that the
difference among the three types of nonwords was
significant by subjects but not by items. Perry et
al. interpreted their results to mean that adults
generally use correspondences at the level of single
phonemes and single graphemes in spelling pro-
duction. These correspondences do not take rime
context into account. Perry et al. found clear ev-
idence for use of rimes in other tasks, such as
judging which of two spellings was more wordlike.
However, their conclusion with regard to spelling
production was that in general, adults use pho-
neme–grapheme sized relationships when spelling
nonwords.
So far, then, the results concerning the rime
constituency hypothesis are mixed. One study of
adults� spelling production (Treiman & Zukowski,
1988) found evidence for this hypothesis but two
studies (Barry & Seymour, 1988; Perry et al.,
2002) did not. Results on this issue have impor-
tant implications for models of spelling. One ap-
proach is a dual-route view (e.g., Barry &
Seymour, 1988; Kreiner & Gough, 1990). In this
view, lexical knowledge and nonlexical knowledge
provide two different routes to spelling. Within
this framework, it is important to determine
whether the nonlexical route is best characterized
in terms of associations between individual pho-
nemes and individual graphemes that are not af-
fected by the phoneme�s context or in terms of
associations that involve larger units and/or that
are sensitive to context. Single-route models,
based on connectionist principles, provide another
potential way to model spelling production (e.g.,
Olson & Caramazza, 1994). Depending on their
structure, such models can handle various types of
contextual effects. Behavioral data on whether
context effects occur and whether rime context is
particularly important will provide essential in-
formation for the development of spelling models.
An understanding of context effects in adult
spelling production should have important impli-
cations, as well, for our understanding of how
children learn to cope with the English writing
system. The system is complex, and simple pho-
neme-to-grapheme rules often yield incorrect
spellings. This is especially true for vowels, which
cause particular difficulty for English learners
(e.g., Treiman, 1993). If adults attempt to cope
with the variability of vowel spelling by learning
how vowels are spelled in particular environ-
ments, when and how does this ability develop?
Some researchers (e.g., Goswami, 1988) have
suggested that children use rime units from an
early age, whereas others (e.g., Frith, 1985) claim
that larger units are used only later in develop-
ment. Information about adults� spelling produc-
tion provides a foundation for studies of children.
Rimes may play a special role in English
spelling and reading because of their status as
phonological units. Alternatively, or in addition,
450 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468
the nature of the English writing system may
make rimes particularly useful. Evidence on this
issue comes from Kessler and Treiman (2001),
who analyzed the sound–spelling relationships in
virtually all of the monosyllabic, monomorphemic
words of English that are familiar to college stu-
dents. They asked whether the spelling of a pho-
neme in one position of the syllable—onset, vowel,
or coda—is significantly affected by the other
constituents of the syllable. The results showed
that sound-to-letter relations are more variable
for vowels than for consonants. However, the
coda has a strong and significant influence on the
spelling of the vowel, with almost all English
vowels showing some associations of this kind.
For example, =e= is more likely to be spelled as ea
when followed by /d/ than when followed by other
consonants. If spellers take advantage of such
associations, vowel spelling becomes more pre-
dictable than it would otherwise be, though by no
means perfectly regular. Kessler and Treiman
found that the onset has a reliable influence on
vowel spelling as well. However, the effect of the
onset on the vowel is weaker overall than the effect
of the coda on the vowel, and onset effects are
only found in some cases.
The results of Kessler and Treiman (2001)
document the patterns of sound-to-spelling asso-
ciation in English monosyllabic words, showing
what patterns are available for use. In the present
study, we ask whether spellers take advantage of
these statistical patterns. Coda-to-vowel associa-
tions were examined in Experiment 1. When
spelling medial vowels in nonwords, do college
students take account of the coda in those cases
for which Kessler and Treiman found a significant
association between vowel and coda? According
to the rime constituency hypothesis, the answer to
the question should be ‘‘yes.’’ Experiment 2 ad-
dressed a similar question for onset-to-vowel as-
sociations. If rimes play a special role in spelling,
as the rime constituency hypothesis claims, then
spellers may not be sensitive to associations that
cross the onset–rime boundary. Experiment 3 ex-
tended the investigation to real words, and Ex-
periment 4 was designed to shed further light on
the linguistic domain over which context has its
effects.
Experiment 1
In this experiment, we asked whether adults�spellings of medial vowels in nonwords are
affected by the identity of the coda. Is the same
vowel transcribed differently when it occurs in
certain contexts—those for which Kessler and
Treiman (2001) documented a significant coda-to-
vowel association—than when it occurs in other
contexts? Such differences are expected according
to the rime constituency hypothesis. Six cases of
coda conditioning were chosen for study. We se-
lected cases in which spelling variations reflect the
identity of the coda (e.g., head vs hem) rather than
the presence versus absence of a coda (e.g., date vs
day).
Table 1 provides information about the six
cases that we investigated. In each case, we iden-
tified two contrasting environments such that a
particular spelling of a vowel, henceforth called
the critical spelling, occurs more often in one case
than the other. Consider Case 1, which involves
the vowel =e=. The critical spelling of this vowel,
ea, is found more often when =e= is followed by
/d/—the experimental context—than when =e= is
followed by /b/, =dZ=, /g/, /k/, /p/, and =tS=—the
control context. In the experimental context, as
Table 1 shows, =e= is spelled as ea in 43% of the
familiar monosyllabic, monomorphemic words
studied by Kessler and Treiman (2001). The con-
trol context was defined so as to minimize the
number of critical spellings, and in this context
=e= is never spelled as ea in the monosyllabic
words analyzed by Kessler and Treiman. Given
the patterns in English monosyllabic words, there
is thus a clear difference between how =e= is
spelled in the experimental context and the con-
trol context. Experiment 1 was designed to de-
termine whether adult spellers are sensitive to this
statistical difference.
Kessler and Treiman�s (2001) analysis of
sound–spelling relationships was limited to
monosyllabic words. All of the nonwords in the
present experiment are monosyllabic, but knowl-
edge derived from experience with polysyllabic
words might affect people�s spelling of monosyl-
lables. To determine whether the statistics are
similar when a larger sample of words is exam-
ined, we examined the spelling of each vowel
phoneme under investigation in a larger sample of
familiar words that included the final syllables of
polysyllables as well as monosyllables. This and
the other reported counts that include polysylla-
bles are based on the words in two machine-
readable dictionaries (Ward, 1993; Weide, 1995);
we eliminated words that are likely to be unknown
to our student population and we counted each
morpheme once, even if it appeared in more than
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 451
one word. As Table 1 shows, the percentage of ea
spellings in the experimental context in the larger
set was 40%, similar to the 43% obtained for
monosyllabic words. In the larger set, again, =e= is
never spelled as ea in the control context.
Once the experimental and control contexts
were defined, we constructed one set of monosyl-
labic nonwords in which =e= appeared in the ex-
perimental context and a matched set of
monosyllabic nonwords in which =e= appeared in
the control context. Sample items are shown in
Table 1. According to the rime constituency hy-
pothesis, spellers should be more likely to use ea
in the experimental nonwords than the control
nonwords.
The logic was similar for the other five cases.
The vowel in Case 2 was /i/. The most common
spelling of this vowel in medial position is ea, as in
beam. In words that end with /d/ and /p/, however,
ee as in reed and deep outnumbers ea. Before
certain other consonants—the control contexts
listed in Table 1—the critical ee spelling is much
less common. In Case 3, the vowel was /o/. In
medial position, it is most often spelled with o
followed by silent final e, as in bone. Before /l/,
however, bare o as in toll and control is fairly
common. Case 4 involved the vowel =u=. When
stressed, it is often spelled as u between a labial
onset and the coda /l/, /s/, =S=, or =tS=. Examples
are full, puss, bush, and butch. Before /f/ and /k/,
oo is the typical spelling, as in book and hoof. The
vowel examined in Case 5 was =au=. This vowel is
usually spelled as ow before final /l/ and /n/, as in
howl and clown. In the control contexts listed in
Table 1, ou is the more common spelling. The final
case was that of =ai=. Before /t/, igh as in bright
and right is fairly common. In the control con-
texts, i followed by final e is more typical.
In most of the cases described above, multiple
spellings for a single phoneme in present-day
English result from a sound change that occurred
after the English spelling system crystallized. For
example, bread and bred were originally pro-
nounced differently, as reflected in the spellings.
Later, a sound change in the former word caused
the two words to be pronounced alike, but the
spellings were not updated. Sound changes have
not always occurred in the same ways in all parts
of the English-speaking world, but for this and the
following experiments we tried to select patterns
that would apply to most dialects of English. In
Case 5 (=au=), the spelling variation does not re-
flect a sound change, but is related to the fact that
=au= is typically spelled as ow in syllable-finalTable
1
Case
inwhich
vowel
spellingis
affected
byco
dain
theword
sofEnglish
Case
1:=e=
Case
2:/i/
Case
3:/o/
Case
4:=u=
Case
5:=au=
Case
6:=a
i=
Followingco
ntextfo
rex
perim
ental
nonword
s
/d/
/d/,
/p/
/l/
=l=;=s=;=S=;=tS=
/l/,
/n/
/t/
Followingco
ntextfo
rco
ntrol
nonword
s
=b=;=d
Z=;=g
=;
=k=;=p
=;=tS=
=ð=;=g
=;=m=;
=st=;=h
==b
=;=d
=;=ð
=;=f=,
/g/,
/k/,
/n/,
/p/,
=t=;=tS=
=f=;=k
==d
Z=;=s=;=tS=
/b/,
/d/,
/f/,
/s/,
/v/,
/z/
Sample
experim
entalnonword
=kled=
=S
r ip=
/dwol/
=juS=
=maul=
=d
r ait=
Sample
controlnonword
=kleb=
=S
r ig=
/dwot/
=juf=
=mautS=
=d
r aib=
Critica
lvowel
spelling
eaee
o,no
followinge
uow
igh
Pro
portion
realword
swith
critical
spelling,ex
perim
entalco
ntexta
.43(.40)
.74(.64)
.36(.33)
.88(.73)
.88(.89)
.54(.39)
Pro
portion
realword
swith
critical
spelling,co
ntrolco
ntexta
.00(.00)
.14(.15)
.00(.00)
.00(.00)
.00(.00)
.00(.00)
aThefirstvalueis
basedonth
efamiliarmonosy
llabic
word
sfrom
Kessler
andTreim
an(2001);th
eseco
ndvalueis
basedonth
efinalsy
llablesoffamiliarword
s,polysy
llabic
aswellasmonosy
llabic.See
textfo
rdetails.
452 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468
position and ou elsewhere. For reasons that are
unclear, the ow spelling also occurs in medial
position before /l/ and /n/.
Method
Stimuli. We constructed 10 pairs of experi-
mental and control nonwords for each of our six
cases. The nonwords in each pair contained the
same onset and the same vowel; they differed only
in the coda. All of the nonwords were phonolog-
ically legal in English. Sample stimuli for each
case are shown in Table 1. In addition to the 60
experimental and 60 control stimuli, there were 20
filler nonwords. The fillers included some vowel
and consonant phonemes that did not appear in
the experimental and control stimuli, and they
had a wider range of syllable structures. The in-
clusion of the fillers increased the variety among
the stimuli and decreased the repetitiveness of the
list. A complete list of the stimuli appears in the
Appendix.
Three different orders were prepared for pur-
poses of presentation. In each order, the experi-
mental items, control items, and fillers were
randomly intermixed with the constraint that no
more than two consecutive items had the same
onset, vowel, or coda.
Procedure. The participants were tested indi-
vidually or in small groups. Each individual or
group was quasi-randomly assigned to one of the
three orders. The participants were told that they
would be asked to spell a series of ‘‘made-up
words.’’ They were asked to pretend that these
were ordinary, everyday words of English and to
spell each item the way they thought it would be
spelled it if were a real word. This instruction was
designed to counteract any tendency people may
have to assume that unfamiliar words are foreign
and should therefore be spelled with atypical
patterns. The experimenter pronounced each item
and the participants repeated it. If a participant
did not repeat an item correctly, the experimenter
said it again and the participant was asked to say
it again. The participants then spelled the item on
prepared answer sheets. A rest break was given
halfway through the list.
Participants. For this and the following ex-
periments, we tested students at Wayne State
University who were native speakers of English
and who reported no history of speech, hearing,
or reading disorders. The students in this and the
following studies participated as volunteers or in
exchange for extra credit. Of the 26 such students
tested for Experiment 1, data from four students
were not analyzed because three or more of their
spellings of the 20 filler items were not plausible
renditions of the item�s phonemic structure. Ex-
amples of these implausible spellings are zaspa for
/væsp/ and swonith for =S c
in=. These students may
have had unacknowledged spelling problems or
may not have been paying full attention to the
task. The latter is a potential problem given that
Experiment 1, unlike the following experiments,
was run at the end of a semester when students are
likely to be tired and overworked. When scoring
filler spellings as plausible or implausible for this
and the following experiments, we considered
spellings that were appropriate in any context to
be plausible. For example, knith for =naið= was
considered plausible because /n/ is spelled as kn in
some words and =ai= is spelled with single i in
some contexts. We accepted reasonable di-
alect-based variations. The 22 participants whose
data were analyzed for Experiment 1 produced a
mean of 0.82 implausible spellings of the 20 filler
items.
Results and discussion
Table 2 shows the mean number of critical
spellings for the vowels in the experimental and
control nonwords. For each vowel, we carried out
t tests by subjects and by items to determine
whether there were significantly more critical
spellings for the experimental items than the
control items. As Table 2 shows, the difference
between experimental and control items was reli-
able by subjects for all cases. By items, the dif-
ference was significant for all but Case 4 (=u=),where it was marginal. For each case, we also
calculated the number of participants and the
number of item pairs for which a difference in the
predicted direction was found. These data, pre-
sented in Table 2, show substantial differences in
the predicted direction for all but Case 4. Given
that certain rimes were repeated across the ex-
periment, we also examined participants� re-
sponses to the first occurrence of each rime. The
proportion of critical vowel spellings was higher
for the experimental items than the control items
in all six cases, and the results were quite similar
to those found in the analysis of all items. Thus, it
appears that adults generally honor coda-to-vowel
associations in their spelling.
The results of Experiment 1 concur with those
of Treiman and Zukowski (1988) in showing that
adults� spellings of vowels are affected by the
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 453
identity of the following consonant. The results do
not support the view that rimes play little or no
role in spelling production and that the weights on
phoneme–grapheme correspondences are derived
from the lexicon as a whole (Barry & Seymour,
1988; Perry et al., 2002). However, our results are
consistent with those of Barry and Seymour and
Perry et al. in showing that the grapheme option
that is overrepresented in a particular rime is not
necessarily produced more often than the gra-
pheme that typically spells the phoneme. With
=ai=, for instance, people produced i plus final e
spellings more often than igh spellings for words
ending in /t/ as well as for other words. Although
our results agree in this respect with those of
Barry and Seymour and Perry et al., our results go
beyond theirs in showing that igh was produced
significantly more often for /aIt/ than for control
rimes. Rime context plays a role in spelling
production.
According to the rime constituency hypothesis,
spellers use associations within the rime unit but
do not go beyond the boundaries of the rime. The
results of Treiman and Zukowski (1988) support
this hypothesis, in that coda-to-vowel but not
onset-to-vowel effects were found in that study.
An alternative view is that adults are sensitive to
the effects of onsets as well as codas on vowel
spelling. The weak results for Case 4 of the present
study could be taken to support this alternative
explanation. This is because the critical oo spelling
of Case 4 tends to occur in real words that have
both a labial onset and a coda /l/, /s/, =S=, or =tS=.Although all of our experimental items had one of
these codas, most did not have a labial onset. That
the results were weaker for Case 4 than for the
other five cases could be taken to suggest that
the onset as well as the coda contributes to the
spelling of the vowel.
Experiment 2 used a design similar to that of
Experiment 1 to determine whether adult spellers
are sensitive to onset-to-vowel associations. Ac-
cording to the rime constituency hypothesis,
spellers do not use onset-to-vowel associations
even when they exist in the writing system because
these associations involve phonemes from differ-
ent constituents of the syllable.
Experiment 2
The purpose of Experiment 2 was to determine
whether adults� spelling of vowels is systematically
affected by the identity of the preceding conso-
nant. Kessler and Treiman (2001) found that as-
sociations between vowels and onsets are less
numerous and less strong overall than associa-
tions between vowels and codas. However, they
identified four cases in which the vowel spelling is
significantly associated with the preceding con-
sonant. The three clearest of these cases were
selected for study here.
The first case of onset conditioning involves
= A=. This vowel is normally spelled as o, as in pod
Table 2
Results of Experiment 1
Case 1: =e= Case 2: /i/ Case 3: /o/ Case 4: =u= Case 5: =au= Case 6: =ai=
Mean (SD) proportion of
critical spellings,
experimental nonwords
.11 (.21) .72 (.31) .23 (.24) .54 (.23) .51 (.24) .27 (.33)
Mean (SD) proportion of
critical spellings, control
nonwords
.05 (.01) .45 (.30) .12 (.12) .43 (.27) .08 (.13) .00 (.00)
p value for difference,
one-tailed t test by subjects
p ¼ :023 p < :001 p ¼ :005 p ¼ :031 p < :001 p < :001
No. of participants (of 22)
showing predicted difference
(+), opposite pattern ()),and tie
7+ 19+ 12+ 11+ 21+ 14+
1) 0) 5) 5) 0) 0)14 tie 3 tie 5 tie 6 tie 1 tie 8 tie
p value for difference,
one-tailed t test by items
p ¼ :039 p < :001 p ¼ :01 p ¼ :08 p < :001 p < :001
No. of item pairs (of 10)
showing predicted
difference (+), opposite
pattern ()), and tie
6+ 9+ 8+ 5+ 10+ 10+
2) 1) 1) 5) 0) 0)2 tie 0 tie 1 tie 0 tie 0 tie 0 tie
454 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468
and slop. However, it is usually spelled as a after
/w/, as in wad and swan. The monosyllables
studied by Kessler and Treiman (2001) do not
contain any exceptions to this latter pattern, and
the association is in this sense stronger than any of
those studied in Experiment 1. As Table 3 shows,
though, there are some exceptions in the first
syllables of the larger set of words that we inves-
tigated for the present study, including wobble and
wombat. Our second case of onset conditioning
involves =g=. After /w/, this vowel is usually
spelled as or, as in work and worth. Were is a
notable exception to this pattern, and there are a
handful of others, such as whirl. After phonemes
other than /w/, =g= does not often take on a
spelling that begins with o. Instead, it has a variety
of spellings, including ur, ir, and er. Ur is partic-
ularly common after /k/, /bl/, and /sl/, as in curb,
blur, and slur, and these onsets were selected as the
control context for the present study. Case 3 in-
volves /u/. When /u/ is preceded by a noncoronal
consonant (i.e., one that does not have a dental,
alveolar, or palato-alveolar place of articulation),
it usually has a spelling beginning with o. The
most common such spelling is oo, as in food and
spoon. Although there are no exceptions to this
pattern in the monosyllables studied by Kessler
and Treiman, some exceptions occur in the larger
set, including scuba and cuckoo. After coronals, u
followed by final e (e.g., rune), eu (e.g., rheum),
and ew (e.g., blew) are more common, although oo
also occurs (e.g., broom).
Two of the patterns described (Cases 1 and 3)
above are due to sound changes that have oc-
curred in many dialects of English. For example,
the effect of context on the spelling of = A= reflects
the fact that the vowel of words like wad and swan
was originally pronounced the same as the vowel
of pad. The influence of the preceding /w/ even-
tually changed the vowel to = A=, the same vowel
as in pod. Given the conservative nature of the
English writing system, the spellings of the vowels
remained the same. The other effect (Case 2) is due
to spelling conventions. English resists placing u
and w (‘‘double u’’) next to one another, and so
vowels that might normally be spelled with u have
a special spelling next to w (e.g., worm instead of
wurm).
We designed pairs of nonwords to investigate
each of the three cases. If adult spellers are sen-
sitive to onset conditioning, the critical vowel
spelling should be more common in the experi-
mental items than the control items.
Method
Stimuli. For each case, 10 pairs of phonologi-
cally legal experimental and control nonwords
were constructed. The experimental and control
syllables in each pair had different onsets but the
same codas. For example, =skw Anz= (experimen-
tal syllable for Case 1) was paired with =slanz=(control syllable). For Case 1, we used codas that
do not condition the a spelling of = A=. Twenty
filler nonwords were constructed. The fillers used
different vowels from the critical stimuli, and they
included a number of different consonants as
well. A complete list of the stimuli appears in the
Appendix.
For purposes of presentation, three different
random orders were prepared. In each list, the
experimental items, control items, and fillers were
intermixed with the constraint that no more than
two consecutive items had the same onset, vowel,
or coda.
Procedure. The procedure was like that of
Experiment 1, except that no rest break was
provided due to the smaller number of stimuli.
Table 3
Cases in which vowel spelling is affected by onset in the words of English
Case 1: = A= Case 2: =g= Case 3: /u/
Preceding context for experimental nonwords /w/ /w/ Noncoronal consonant
Preceding context for control nonwords Any other consonant /bl/, /k/, /sl/ Coronal consonant
Sample experimental nonword =skw Anz= =wgdZ= /muk/
Sample control nonword =sl Anz= =kgdZ/ =p ruk=
Critical vowel spelling a initial o initial o initial
Proportion real words with critical spelling,
experimental contexta1.00 (.86) .78 (.67) 1.00 (.87)
Proportion real words with critical spelling,
control contexta.09 (.11) .00 (.15) .49 (.30)
a The first value is based on the familiar monosyllabic words from Kessler and Treiman (2001); the second value is
based on the first syllables of familiar words, polysyllabic as well as monosyllabic. See text for details.
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 455
The participants were assigned to one of the three
lists in the order in which they were run.
Participants. Twenty-one students contributed
data. The data of one additional participant were
dropped because more than two of this student�sspellings of the 20 filler items were not plausible
renditions of the item�s phonemic structure. The
other participants produced a mean of 0.42 im-
plausible spellings of the filler items.
Results and discussion
For Case 1 ð= A=Þ, we tabulated the number of
vowel spellings that began with a for the experi-
mental syllables and the control syllables. Such
spellings should be more frequent for experimen-
tal syllables such as =skw Anz= than for control
syllables such as =sl Anz= if spellers are sensitive to
the influence of the preceding consonant on the
spelling of the vowel. Spellings of the vowel with
au or aw were not counted as critical vowel
spellings, since these spellings could reflect a dia-
lect in which = A= has merged with = c=, which is
normally spelled as au or aw. Such spellings were
uncommon, however, occurring less than 4% of
the time. As Table 4 shows, the critical spellings
were substantially more common for the experi-
mental stimuli than the control stimuli, and the
difference was significant according to t tests by
subjects and by items. All participants showed a
difference in the predicted direction, and all pairs
of experimental and control nonwords did as well.
For Case 2 (=g=), the critical vowel spellings
were those that began with o. As Table 4 shows,
such spellings were significantly more frequent for
the experimental nonwords such as =wgdZ= than
for the control nonwords such as =kgdZ=. Al-
though o spellings were more common for the
experimental stimuli than the control stimuli, u
was the most common type of spelling for both
types of items. E was used almost as often as o for
the experimental nonwords (28% e vs 29% o), but
e was less common for the control nonwords
(15%). Spellers� use of e after /w/ may reflect the
influence of were. The most important result,
though, is that o spellings were significantly more
common for experimental than control nonwords.
For Case 3 (/u/), the critical vowel spellings
were those that began with o. These spellings were
much more common for the experimental non-
words such as /muk/ than for the control non-
words such as =p ruk=. In the great majority of
cases, the particular o-initial grapheme that was
used was oo.
Given that certain onset-vowel sequences were
repeated across the experiment, we examined
participants� spellings for the first occurrence of
each sequence. In all three cases, the proportion of
critical vowel spellings was higher for experimen-
tal nonwords than for control nonwords. The re-
sults were numerically similar to those observed
for all items.
Thus, there is strong evidence that adults�spellings of vowels are influenced by the preceding
consonant. This evidence was found for all three
of the cases that we examined. These results in-
dicate that adults pick up and use onset-to-vowel
associations when they exist in English. These
findings fail to support the rime constituency hy-
pothesis, which predicts that spellers use associa-
tions within the syllable rime but not those that
involve the onset and part of the rime. Under
General Discussion, we consider why the adults of
Experiment 2 were influenced by onsets in the
spelling of vowels, whereas those in the study of
Treiman and Zukowski (1988) were not.
Table 4
Results of Experiment 2
Case 1: = A= Case 2: =g= Case 3: /u/
Mean (SD) proportion of critical spellings,
experimental nonwords
.87 (.20) .29 (.31) .87 (.13)
Mean (SD) proportion of critical spellings, control
nonwords
.15 (.13) .01 (.02) .29 (.17)
p value for difference, one-tailed t test by subjects p < :001 p < :001 p < :001No. of participants (of 21) showing predicted 21+ 15+ 21+
difference (+), opposite pattern ()), and tie 0) 0) 0)0 tie 6 tie 0 tie
p value for difference, one-tailed t test by items p < :001 p < :001 p < :001
No. of item pairs (of 10) showing predicted 10+ 10+ 10+
difference (+), opposite pattern ()), and tie 0) 0) 0)0 tie 0 tie 0 tie
456 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468
The findings of Experiment 2 are surprising in
light of the view that rimes play a special role in
reading and phonological processing in English.
Before considering how to explain these findings,
we thought it important to determine whether the
same sensitivity to onsets appears in a spelling
task involving real words. Experiment 3 was de-
signed to address this issue. It also asked whether
the coda influences adults� spellings of vowels in
real words.
Experiment 3
To determine whether people use consonantal
context when spelling the vowels of real words, we
can look at their errors. If spellers store only in-
formation about the most common spellings of
vowels, then a word like wombat would have the
expected vowel spelling. This is because it uses the
typical grapheme for medial = A=, o. Possum, too,
has the expected vowel. In this view, substitutions
of a for o should be uncommon for both wombat
and possum. However, if spellers store informa-
tion about the spellings of vowels after particular
consonants, then wombat would have an irregular
vowel spelling. This is because the most common
spelling of = A= after /w/ is a. In this view, substi-
tutions of a for o should be relatively common for
wombat, more common than for possum. We
asked whether substitution errors differ in this
way for experimental words like wombat and
control words like possum.
Experiment 3 used the logic described above to
examine both onset-to-vowel and coda-to-vowel
associations. For each case of coda-to-vowel as-
sociation (Table 1) and each case of onset-to-vo-
wel association (Table 3), we selected real words
for which the vowel is not spelled in the critical
manner. Some of these words had the vowel in the
experimental context and an equal number had
the vowel in the control context. If context has an
effect on spelling, then substitutions of the critical
spelling for the correct spelling should be more
common for the experimental words than the
control words.
Method
Stimuli. For each of the cases examined in
Experiment 1 and Experiment 2, we chose as
many words as possible that contained the vowel
in the experimental context and where the vowel
was not spelled in the critical manner. We selected
an equal number of such words that contained the
vowel in the control context. We attempted to
equate the experimental and control words for
frequency, length, stress of the critical vowel, and
conventional spelling of the critical vowel. We
avoided words that had homophones that might
be familiar to our students. Also, we tried to
avoid situations in which use of the critical vowel
would result in a real word. Because the number
of potential stimuli was relatively small, we could
not use an equal number of stimuli in all of the
cases.
For the test of coda-to-vowel associations, the
number of stimuli per case ranged from a low of 2
(1 experimental word and 1 control word) for
Case 4 to a high of 20 (10 experimental words and
10 control words) for Case 3. There were 30 ex-
perimental words and 30 control words. Across
the six cases, the experimental words and control
words were alike in length (mean of 6.2 letters for
both). The experimental words were slightly more
frequent than the control words (33.5 and 28.8,
respectively, according to Zeno, Ivenz, Millard, &
Duvvuri, 1995), but the difference in frequency
was not significant ðp ¼ :75Þ.For the test of onset-to-vowel associations, the
number of stimuli per case ranged from 4 (2 ex-
perimental and 2 control) to 12 (6 experimental
and 6 control). Altogether, there were 12 experi-
mental words and 12 control words testing onset-
to-vowel associations. The experimental and
control words were similar in length (5.5 letters vs
5.4 letters). The experimental words were slightly
but not significantly more frequent than the con-
trols (25.3 vs 19.9; p ¼ :58). All of the words used
in the experiment are listed in the Appendix.
For purposes of presentation, the 84 words
were randomly intermixed with the constraint that
no two words with the same critical vowel were
adjacent. Two different random orders were pre-
pared.
Procedure. The participants were tested indi-
vidually or in small groups. Approximately half of
the participants received the stimuli in each of the
two random orders. The experimenter pro-
nounced each word, said it in a sentence, and then
said the word again. The participants were then
asked to spell the word on prepared answer sheets.
There was a rest break halfway through the list.
Participants. The participants were 35 students
from the same population as in the previous ex-
periments, with the exception that one participant
was a student at a different college in the same
area.
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 457
Results and discussion
The participants generally found the words
difficult to spell. The mean overall error rate was
27.6% (range 2.4–66.7%). Our primary interest
was in the types of spelling errors made for ex-
perimental words and control words. Specifically,
were participants more likely to substitute the
critical vowel for the correct vowel when spelling
the experimental words than when spelling the
control words? If so, this would indicate that
adults show an effect of context on spelling.
For the coda-to-vowel experimental words,
22% of the errors used the critical vowel spelling
in place of the conventional spelling. For example,
participants sometimes misspelled shred as shread
and casserole as casserol. For the coda-to-vowel
control stimuli, only 8% of the errors used the
critical vowel. That is, errors such as fleak for fleck
and isotop for isotope were relatively uncommon.
As Table 5 shows, the proportion of errors that
used the critical vowel spelling was reliably greater
for the experimental words than the control
words, both by subjects and by items. The re-
maining errors on the experimental and control
words misspelled phonemes other than the target
vowel, as in cassarole and caserole for casserole. In
a few cases, the target vowel was spelled with a
grapheme other than the critical vowel, as in flak
for fleck.
With the onset-to-vowel stimuli, too, spellers
were more likely to use the critical spelling for the
experimental words than the control words. For
the experimental words, 33% of the errors sub-
stituted the critical vowel spelling for the correct
spelling. For example, wok was sometimes spelled
as wak and wombat was sometimes spelled as
wambat. The control stimuli were less likely to
give rise to such errors. For these stimuli, just 5%
of the errors used the critical vowel spelling. Cog
was rarely spelled as cag, for example, and possum
was rarely spelled as passum. Statistical tests,
shown in Table 5, confirmed that the experimental
and control words differed reliably in the pro-
portion of critical errors relative to all errors.
Other errors made by participants misspelled
phonemes other than the target vowel, as in
whombat for wombat, or misspelled the target
vowel with a grapheme other than the critical
spelling, as in possome for possum. The errors were
typically nonwords (86% nonwords pooling over
onset-to-vowel and coda-to-vowel stimuli). Given
that some dictionaries list coocoo as a slang al-
ternative for cuckoo, we also analyzed the results
without this item. The difference between experi-
mental and control words in the proportion of
critical errors relative to all errors was still sig-
nificant (by subjects: p ¼ :001, one-tailed; by
items: p ¼ :023, one-tailed).In light of the small numbers of stimuli for
some of the individual vowels, statistical tests were
not carried out for each case separately. However,
the proportion of critical errors relative to all er-
rors was higher for experimental words than for
control words for Cases 1, 2, 3, 5, and 6 of the
coda-to-vowel stimuli and Cases 1–3 of the onset-
to-vowel stimuli. The only case that did not show
a difference in the predicted direction was coda-to-
vowel Case 4, =u=. There was just one experi-
mental word and just one control word for this
case, however. The experimental word, swoosh,
may not be an ideal test because the critical
spelling, u, rarely appears next to w in English.
Because some of the critical rimes and onset-
vowel sequences were repeated within the list, we
examined participants� performance on the first
Table 5
Results of Experiment 3
Coda-to-vowel Onset-to-vowel
Proportion (and number) critical errors relative to all errors,
experimental words
.22 (49/225) .33 (54/162)
Proportion (and number) critical errors relative to all errors,
control words
.08 (24/269) .05 (8/155)
p value for difference in proportions, one-tailed t test by subjects p ¼ :004 p < :001
No. of participants (of 35) showing predicted 21+ 26+
difference (+), opposite pattern ()), and tie 7) 1)6 tiea 6 tieb
p value for difference, one-tailed t test by items p ¼ :048 p ¼ :012
aOne additional participant made no errors on coda-to-vowel items.bOne additional participant made no errors on onset-to-vowel items, and another made no errors on onset-to-vowel
experimental items.
458 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468
instance of a given sequence that they encoun-
tered. The proportion of errors that used the
critical vowel spelling was larger for the experi-
mental items than the control items for both the
coda-to-vowel case and the onset-to-vowel case,
with the results numerically very similar to those
observed for all items.
Participants who were more sensitive to coda-
to-vowel associations (as measured by the differ-
ence in proportion of substitution errors of the
critical vowel on the experimental and control
words) also tended to be more sensitive to onset-
to-vowel associations (r ¼ :61; p < :001, one-
tailed; the results of three participants could not
be included in this analysis because they made no
any errors in one or more of the cells). A com-
bined measure of sensitivity to coda-to-vowel and
onset-to-vowel associations showed a significant
correlation with performance on the spelling test
such that better spellers showed a larger difference
between experimental and control items than did
poorer spellers (r ¼ :29; p ¼ :048, one-tailed; all
participants contributed data to this analysis).
These results suggest that good spellers at the
college level are more responsive to the statistical
patterns that affect vowel spelling than are poorer
spellers.
Are spellers more sensitive to those patterns
for which stronger evidence exists within the En-
glish spelling system itself? Relatively few spelling
patterns were examined in each experiment, and
few examples of some of the patterns were in-
cluded in Experiment 3. Nevertheless, we carried
out exploratory analyses to address this question.
For Experiment 3, we expressed spellers� sensi-
tivity to each pattern in terms of the difference
between experimental and control words in the
proportion of critical errors relative to all errors.
For Experiments 1 and 2, we used the difference
between experimental and control items in pro-
portion of critical spellings. We did not include
the results for coda-to-vowel Case 4, since this
case may also tap onset-to-vowel sensitivity, as
mentioned above. For the combined data from
Experiments 1 and 2, there was a significant
Spearman rank-order correlation between the
strength of the behavioral effect and the strength
of the pattern in the English writing system when
the latter was expressed as the difference in the
proportion of real words with the critical spelling
in the experimental and control contexts. This
held true whether the pattern�s strength was cal-
culated on the basis of the monosyllabic words
from Kessler and Treiman (2001) or on the basis
of the larger word set that was analyzed for the
present study (rs ¼ :73; p ¼ :020, one-tailed, for
the former analysis; rs ¼ :84; p ¼ :005, one-tailed,for the latter analysis). For Experiment 3, where
there were generally fewer items for each case, a
significant rank-order correlation was found when
we used the data from the larger word set
(rs ¼ :71; p ¼ :023, one-tailed). The correlation
did not reach significance using the smaller word
set (rs ¼ :55; p ¼ :080, one-tailed). These results
must be interpreted with caution, as only eight
cases were included in the analyses and because
there are questions about the best way to assess the
strength of a pattern in the writing system.
However, the findings provide initial support for
the idea that spellers are especially sensitive
to those patterns for which strong evidence exists
in the words to which they have been exposed.
To summarize, the results of Experiment 3
support the conclusions drawn from Experiments
1 and 2 and show that the findings generalize to
real words. Adults are sensitive to the effects of
both onset and coda on the spelling of medial
vowels. They use this information when spelling
real words as well as when constructing spellings
for nonwords. With real words, this sensitivity
causes certain vowel substitutions to be more
common in some consonantal contexts than in
other contexts. The results of Experiment 3 do not
support the theory that spellers store a list of the
grapheme options for each vowel that is weighted
by the overall frequency of the graphemes in all
consonantal contexts (Barry & Seymour, 1988).
For some vowels, the weightings vary as a func-
tion of not only the following consonant but also
the preceding consonant.
The results obtained so far provide strong ev-
idence against the rime constituency hypothesis,
according to which spellers notice and use asso-
ciations involving phonemes within the rime but
do not go beyond this domain. An alternative
hypothesis, which is compatible with the results of
Experiments 1–3, is that linguistic boundaries are
not critical at all. In this view, spellers pick up
context-sensitive associations involving adjacent
phonemes and generalize these associations in a
way that is blind to linguistic boundaries. We may
call this hypothesis the adjacency hypothesis. An-
other possibility, which is also compatible with the
results of Experiments 1–3, is that spellers are
sensitive to syllable boundaries in their learning
and use of context-sensitive associations. They do
not go beyond the syllable when applying patterns
like those studied here. This latter hypothesis may
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 459
be called the syllable hypothesis. We cannot tease
apart the adjacency hypothesis and the syllable
hypothesis based on the results of Experiments 1–
3 because the relevant context was always in the
same syllable as the vowel in these experiments.
Experiment 4 was designed to distinguish be-
tween the adjacency hypothesis and the syllable
hypothesis. In this experiment, we took a closer
look at coda-to-vowel associations and their ef-
fects on nonword spelling. Consider the associa-
tion between /i/ and following /d/ and /p/ that was
examined in Experiments 1 and 3. In English, /i/
tends to be spelled as ee before /d/ and /p/. It is
more likely to be spelled as ea before other con-
sonants. The results of Experiments 1 and 3 sug-
gest that spellers are sensitive to this association.
However, the stimuli of Experiments 1 and 3 al-
ways had the /i/ and the following /d/ or /p/ within
the same syllable. It is thus not clear whether
membership in the same syllable is critical. To
determine whether it is, Experiment 4 examined
participants’ spelling of /i/ in disyllabic nonwords
such as =S ri0dok= (different syllable, experimental)
and =S ri0gok= (different syllable, control). In syl-
lable division tasks, English-speaking adults gen-
erally place the syllable boundary after the /i/ and
before the following consonant in such cases
(Treiman & Danis, 1988b). This outcome is con-
sistent with linguistic theories of syllabification
according to which the onsets of stressed syllables
are maximized (e.g., Selkirk, 1982). According to
the syllable hypothesis, the syllable boundary
blocks /d/ from selecting ee in =S ri0dok= and ee
spellings should be no more common in =S ri0dok=than in =S ri0gok=. In contrast, the adjacency hy-
pothesis states that the syllable boundary is not
critical. By this hypothesis, the rate of ee spellings
should be higher in =S ri0dok= than in =S ri0gok= even
though the vowel and the following consonant do
not belong to the same syllable in =S ri0dok=.Experiment 4 also included stimuli such as
=S rid0bok= (same syllable, experimental) and
=S rig0bok= (same syllable, control). With such
items, the /i/ and the following /d/ or /g/ are vir-
tually always placed in the same syllable in syl-
labification tasks (Treiman & Zukowski, 1990).
This outcome is consistent with linguistic theories
stating that clusters such as /db/ and /gb/ cannot
belong to the same syllable (e.g., Selkirk, 1982).
The adjacency hypothesis and the syllable hy-
pothesis both predict a higher rate of ee spellings
for =S rid0bok= than for =S rig0bok=. Such a result
would show that the findings of Experiment 1
with monosyllabic nonwords generalize to the first
syllables of disyllabic nonwords. This result would
also forestall the interpretation that a lack of
difference between =S ri0dok= and =S ri0gok= is due to
polysyllabicity rather than to syllable boundaries.
Experiment 4
Method
Stimuli. We selected three of the vowels that
were used in Experiment 1 and constructed di-
syllabic nonwords in which these vowels appeared
in the first syllable and the second syllable was
stressed. There were four types of nonwords,
which varied in position of the syllable boundary
(target vowel and following consonant in same
syllable vs different syllable) and the identity of
the consonant that followed the target vowel
(experimental vs control). To illustrate, consider
the vowel /i/ (Case 2 of Experiment 1). In
=S rid0bok= (same syllable, experimental), this vo-
wel and the following consonant, /d/, are in the
same syllable. In =S rig0bok= (same syllable, con-
trol), the syllable boundary is also between the
two consonants and so /i/ and /g/ are in the same
syllable. In =S ri0dok= (experimental, different syl-
lable) and =S ri0gok= (control, different syllable),
the middle consonant is the onset of the second
syllable. The syllable hypothesis predicts an in-
teraction between context and boundary such that
critical spellings of the vowel are more common in
the experimental context than the control context
in the same-syllable case but not in the different-
syllable case. The adjacency hypothesis predicts a
main effect of context—more critical spellings for
experimental nonwords than for control non-
words—but no interaction with boundary.
We checked whether there are any English
monomorphemic words that our participants
might know that contain the target vowel in an
environment that is very similar to that of the
Experiment 4 stimuli (i.e., target vowel followed
by one or more consonants, followed by a stressed
vowel word-finally). For /i/, the only such words
are clich�ee and Peking, neither of which contains
the critical ee spelling. For =au=, there are no such
words. There are a handful of such words for =ai=,including Taiwan and ideal, but none that use the
critical igh spelling and only one (typhoon) in
which the consonant that follows the vowel is one
of the consonants used in Experiment 4. Thus, if
participants spell =S rid0bok= as shreedboke or
=Saul0wost= as showlwost, their ee and ow spellings
460 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468
cannot reflect experience with highly similar di-
syllabic words. Instead, such spellings would ap-
pear to reflect generalizations from participants�experience with words such as reed and howl. The
/o/ vowel that was used in Experiment 1 (Case 3)
was not used in Experiment 4 because a number
of disyllabic words that are very similar to the
Experiment 4 stimuli use the critical single o
spelling, such as Colette and hotel. Also, it could
be difficult to differentiate the single o spelling
from the o plus final e spelling in the first syllables
of disyllabic words. Because long or diphthon-
gized vowels appear to be critical for producing
the syllabifications desired here (e.g., Treiman &
Danis, 1988b), we could not use =e= (Case 1 of
Experiment 1) or =u= (Case 4 of Experiment 1) in
Experiment 4.
For each of the three vowels used in Experi-
ment 4, we constructed 8 stimuli of each of the
four types. In addition, there were 20 filler non-
words. The fillers were more complex than those
of the previous experiments in that 12 were
monosyllabic and the other 8 were disyllabic. The
disyllabic fillers had stress on the first syllable so
as to vary the stress patterns of the stimuli in the
experiment. The stimuli are listed in the Appendix.
Three different random orders were prepared
for purposes of presentation. In each, the test and
filler nonwords were randomly intermixed such
that no more than two consecutive stimuli had the
same critical vowel and no more than two con-
secutive stimuli were of the same type.
Procedure. The procedure was like that of
Experiment 1. However, given the complexity of
the nonwords, the experimenter pronounced each
nonword twice before asking participants to re-
peat it. The participants were assigned to one of
the three random orders in the order in which they
were run.
Participants. Twenty-three participants con-
tributed data. The results of four additional stu-
dents were dropped because they produced three
or more implausible spellings of the filler items.
The remaining students produced a mean of 0.87
implausible spellings of the 20 fillers.
Results and discussion
Table 6 shows the mean proportions of critical
spellings of each vowel in the four types of non-
words. Analyses of variance were carried out by
subjects and by items using the factors of case,
context (experimental vs control), and bound-
ary type (same syllable vs different syllable).
There was a main effect of case [F 1ð2; 44Þ ¼16:06; F 2ð2; 21Þ ¼ 48:91; p < :001 for both] such
that critical vowel spellings were most common
for /i/, next most common for =au=, and least
common for =ai=. This is the same pattern that
was observed in Experiment 1, although the pro-
portion of critical spellings was in each case lower
than for the monosyllabic nonwords of Experi-
ment 1. The main effect of context was also sig-
nificant [F 1ð1; 22Þ ¼ 17:91; F 2ð1; 21Þ ¼ 16:05;p � :001 for both]. Critically, this main effect was
qualified by an interaction between boundary and
context ½F 1ð1; 22Þ ¼ 16:82; p < :001; F 2ð1; 21Þ ¼11:90; p ¼ :002�. The interaction arose because
there were significantly more critical spellings in
the experimental same-syllable case than in the
control same-syllable case [t1ð22Þ ¼ 5:06; t2ð23Þ¼ 5:26; p < :001, one-tailed for both]. Indeed, 22
of the 23 participants showed a difference in this
direction, as did 19 of the 24 item pairs. In the
different-syllable case, in contrast, there were not
significantly more critical spellings in the experi-
mental context than the control context. The
patterns were similar when we looked at the re-
sults for the first-presented item in each quadru-
plet.
The ANOVAs also showed an interaction
between case and boundary [F 1ð2; 44Þ ¼10:44; F 2ð2; 21Þ ¼ 14:21; p < :001 for both]. For
/i/ and =ai=, there were generally more critical
spellings in the same-syllable cases than in the
different-syllable cases. For =au=, there were more
critical spellings in the different-syllable case.
This is likely because =au= is a syllable-final vowel
in the different-syllable case, and this vowel is
more likely to be spelled as ow when it ends a
syllable (e.g., how and power) than when it is
Table 6
Mean and standard deviation of proportion of critical spellings in Experiment 4
Case 2: /i/ Case 5: =au= Case 6: =ai=
Experimental, same syllable .52 (.34) .25 (.23) .09 (.22)
Control, same syllable .35 (.29) .10 (.18) .00 (.00)
Experimental, different syllable .34 (.32) .28 (.30) .01 (.03)
Control, different syllable .30 (.31) .29 (.29) .01 (.03)
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 461
followed by a consonant in the same syllable (e.g.,
mouse).
The finding that participants were more likely
to use the critical vowel spelling in the experi-
mental same-syllable case than in the control
same-syllable case indicates that the results of
Experiment 1 generalize to the first syllables of
disyllabic items with second-syllable stress. These
patterns are not restricted to monosyllabic items
like those of Experiment 1. The obtained differ-
ences do not seem to depend on the items being
highly similar to real words in both segmental
pattern and stress pattern, for the items of Ex-
periment 4 were not very similar to existing words.
The most important result of Experiment 4 is
that spellers appear to use patterns that involve
the syllable rather than patterns that involve mere
adjacency. For example, /i/ and /d/ must be in the
same syllable for an increased number of ee
spellings to be found. Likewise, igh spellings of
=ai= almost never occur unless =ai= and /t/ are in
the same syllable. For =au=, ow spellings seem to
occur in two main circumstances—when =au= is
followed in the same syllable by /l/ or /n/ or when
=au= is at the end of a syllable. For all three
vowels that we investigated, therefore, adjacency
between a vowel and a following consonant does
not suffice for the critical context-conditioned
vowel spelling to occur. What is important is that
the vowel and the following consonant belong to
the same syllable. These results support the syl-
lable hypothesis over the adjacency hypothesis.
Ideally, one would like to disentangle the fac-
tors of adjacency and membership in the same
syllable for contextual associations involving
vowels and preceding consonants as well as for
those involving vowels and following consonants.
This would be difficult to do, however. For ex-
ample, the consonant that conditions several of
the changes examined in Experiment 2, /w/, can-
not occur in syllable-final position in English.
General discussion
Vowels are difficult to spell in English. How
can one choose the correct spelling of a vowel
from among the many possibilities that are offered
by the writing system? This problem provides a
good test case for questions about how language
users deal with irregularity. In the present exper-
iments, we asked how experienced spellers attempt
to solve the vowel problem by examining their
spellings of vowels in nonwords and their subst-
itution errors on vowels in real words. We asked
two specific questions. First, do spellers use the
consonantal context to help narrow the range of
possible spellings for a vowel? Second, what type
of context do spellers use? The answers to these
questions have implications for theories of skilled
spelling production, views of spelling develop-
ment, and other issues.
In answer to first question, we found that
adults� spellings of syllable-medial vowels are af-
fected by the identity of the surrounding conso-
nants. This result suggests that spellers do not
simply store a list of the possible spellings of each
vowel weighted by the frequency of the phoneme-
to-grapheme correspondence in the language as a
whole (Barry & Seymour, 1988). Nor are the
weights affected only by frequency in that partic-
ular position of the syllable, the factor that
Goodman and Caramazza (1986) and Sanders
and Caramazza (1990) examined in studies of
patients whose spelling ability was compromised
by brain damage. In some cases, our results indi-
cate, the graphemes are weighted by their fre-
quency in a specific consonantal environment.
The results thus suggest that spellers use context
as a way of dealing with the difficulties posed by
the English writing system. The findings of Kess-
ler and Treiman (2001) show that the system be-
comes more reliable, statistically, when context is
taken into account. The present results add to this
picture by showing that experienced spellers use
the statistical patterns. Even with these patterns,
the English spelling system is less regular than
many other alphabetic systems. However, the
patterns help to rationalize some spellings that
would otherwise have to be learned by rote.
The second question that motivated the pres-
ent study dealt with the types of context used by
spellers of English. The results of Experiments 1–3
show that the relevant context for a medial vowel
includes consonants from the same constituent of
the syllable—the coda—and consonants from a
different constituent of the syllable—the onset.
However, the results of Experiment 4 indicate that
the following consonant must be within the same
syllable to have an effect, at least for the patterns
studied here. That is, spellers appear to extend the
associations that they have learned in a way that is
sensitive to syllable boundaries. Syllables may
play a role in English spelling if people segment
to-be-spelled items into syllables and spell one
syllable at a time. A slightly different interpreta-
tion is that spellers consider only phonemes in the
same syllable when deciding how to transcribe a
462 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468
target phoneme. To our knowledge, the present
findings are the first to show a role for syllables in
the spelling of normal adults. However, several
studies have pointed to a role for syllables or
syllable-like units in the recognition of printed
words (e.g., Mewhort & Beal, 1977; Prinzmetal,
Treiman, & Rho, 1986).
Our finding that vowel spelling is influenced by
the onset as well as the coda is surprising on some
interpretations of the view that rimes play a
special role in phonological processing and read-
ing. The present results suggest that people pick
up onset-to-vowel associations when these asso-
ciations exist and that they are as sensitive to
them as to coda-to-vowel associations. To the
extent that rimes play a special role in English
spelling, this may be because associations between
vowels and codas are more numerous in this
writing system than associations between vowels
and onsets (Kessler & Treiman, 2001). This in-
terpretation is consistent with the conclusions of
Martensen, Maris, and Dijkstra (2000), who
found that consideration of the rime unit does not
improve spelling-to-sound predictability in Dutch
and that this unit does not play a special role for
Dutch readers. Martensen et al. suggested that
the characteristics of spelling-to-sound relation-
ships in a particular writing system shape the
units that readers use. Our findings suggest that
the same is true for spelling. In English, spellers
take advantage of associations between onsets
and vowels when these exist. However, associa-
tions between vowels and codas are more influ-
ential overall because the spelling system has
more associations between vowels and codas than
between vowels and onsets. The same may be true
for English reading, but this remains to be
examined.
Recall that Treiman and Zukowski (1988) did
not find a sensitivity to onsets in the spelling of
vowels. For example, their participants rarely
spelled =f reh= as frieth on the basis of friend. Also,
frieth was not rated as a particularly good spelling
of =f reh=. How can we reconcile the lack of an
onset-to-vowel effect in the Treiman and Zukow-
ski study with the substantial onset-to-vowel effect
observed here? An answer is suggested by the fact
that Treiman and Zukowski investigated very
unusual vowel spellings. For example, the ie
spelling of =e= appears only in the word friend.
The onset-conditioned vowel spellings investi-
gated in the present study occur more widely. For
example, = A= is written as a in wad, wan, swan,
swatch, and so on. The present results suggest that
adults are sensitive to onset-to-vowel associations
that appear in a number of different words. In
such cases, they are willing to extend the patterns
to items they have not encountered before. The
results of Treiman and Zukowski suggest, in
contrast, that spellers do not develop generaliz-
able knowledge of onset-to-vowel associations
from single words with unusual spellings like
friend. Having observed that associations between
vowels and codas are more common in English
than associations between vowels and onsets,
adults may sometimes generalize a coda-to-vowel
association from a single word. A single word may
not provide a basis for a generalizable onset-to-
vowel association. When better evidence for an
association between onset and vowel exists, as
with the use of a for = A= in wad, wan, swan,
swatch, and so on, spellers are not restricted by the
onset–rime boundary. This interpretation is con-
sistent with the view described above—that the
role of the rime in English spelling derives from
the fact that the writing system includes more
associations between vowels and codas than be-
tween vowels and onsets.
Our results demonstrate context effects on the
spelling of English vowels, and they show that
these effects can involve the preceding consonant
as well as the following one. An issue that we have
not yet considered is whether the observed con-
textual effects operate on sound-to-letter corre-
spondences or on letters alone. Consider the fact
that =e= has several spellings, including ea and e,
and that participants in Experiment 1 used ea a
higher proportion of the time before /d/ (or its
spelling d) than before /p/ (or its spelling p). We
have assumed that the choice between ea and a
takes into account the phoneme that is being
spelled. However, an alternative hypothesis is that
spellers favor the letter sequence ea more before d
than before p, regardless of the phonemes being
represented. Such a preference could arise if the
proportion of ea spellings in English text is higher
before d than before p and if spellers adjust their
own productions to reflect this. According to this
alternative hypothesis, which we will call the letter
frequency hypothesis, participants may generate
possible spellings of =e= without regard for con-
text and then choose the spelling that yields a
more typical letter sequence. Letter frequency
could well be influential, especially because these
experiments look at a complete spelling that takes
some time to produce.
Our experiments were not designed to tease
apart the two hypotheses discussed above.
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 463
However, we conducted a post hoc check by ex-
amining the results for those cases in which the
distributions of spellings in texts do not fit the
patterns expected by the letter frequency hypoth-
esis. For each test vowel in Experiments 1 and 2,
such as =e=, we considered each pair of experi-
mental and control contexts, such as /d/ and /p/.
We chose a priori the expected spellings in those
environments, using the critical vowel spelling and
the other most likely vowel spelling. For each of
the four spellings—ead, eap, ed, and ep in this
example—we counted how often they appear at
the appropriate edges of words in English text,
based on Zeno et al. (1995). Then we computed
whether the proportion of critical vowel spellings
(ea) was higher in the experimental environment
(d) than the control environment (p). If so, the
letter frequency hypothesis could potentially ex-
plain the experimental results. For 68 of the 90
experimental-control pairs in Experiments 1 and
2, the pattern in text fit the letter frequency hy-
pothesis and so the results for these pairs cannot
help in disentangling the two hypotheses. In 22 of
the cases, however, the pattern in text was con-
trary to the letter frequency hypothesis. For =e=,for instance, the proportion in text of ea spellings
out of both ea and e spellings is .05 before d but
.23 before p. For these cases, we looked at re-
sponses that used one of the four predicted
spellings. There were significantly more critical
vowel spellings in the experimental cases than the
control cases [tð21Þ ¼ 4:24; p < :001, one-tailed].
Because this result obtains even for cases where
the frequency of the letter patterns should influ-
ence participants in the opposite direction, some
of the effects in the present study must operate at
the level of sound-to-letter correspondences.
In other instances, the text counts align with
the letter frequency hypothesis and so the
observed effects may operate at the letter
level. Consider the experimental-control pair
=gait=–=gaif= (Experiment 1, Case 6). Word-final
ight is a more common letter sequence than word-
final ite, whereas final ife is more common than
ighf. Indeed, no English word ends with ighf. The
unusual nature of the ighf sequence is sufficient to
explain why the participants in Experiment 1
avoided it. However, a simple view of letter fre-
quency is not sufficient to explain the results of
Experiment 4. Participants in this experiment used
ight more often than ighf when the corresponding
phonemes were in the same syllable, but they did
not use ight more often than ighf when the pho-
nemes were in different syllables. The notion of a
syllable appears to be necessary in order to ex-
plain these results.
Models of the spelling process are less fully
developed and less explicit than models of the
reading process. The present findings provide
some constraints on the form that spelling models
should take and some data that they will need to
explain. One issue that modelers confront is the
architecture of the spelling system. Researchers
such as Barry and Seymour (1988) and Kreiner
and Gough (1990) have espoused a dual-route
view of spelling, similar to dual-route models of
reading. In this view, lexical knowledge and non-
lexical knowledge provide two different routes to
spelling. Within this framework, our results sug-
gest that the mappings used by the nonlexical
route consider the consonantal context in which a
vowel phoneme occurs. They do not represent
only the most common spelling of each phoneme
or the spelling that is most common in that par-
ticular position of the syllable. Single-route mod-
els, based on connectionist principles, are another
possible architecture (e.g., Olson & Caramazza,
1994). To the extent that such models pick up the
statistical regularities in the writing system, they
may account well for human performance. Single-
route models have the potential to explain evi-
dence pointing to a role of analogies in spelling
(e.g., Campbell, 1983). For example, the =g rol= of
Experiment 1 is similar to the real word =t rol=(troll) and people might therefore spell it in a
similar way. However, the disyllabic stimuli of
Experiment 4 were not highly similar to any real
words and yet the same-syllable items in this ex-
periment showed the same pattern of results
found in the experiments with real words. A
simple analogy-based model would have difficulty
explaining the results of that experiment. This and
other models of spelling need to be more fully
developed before they can be clearly distin-
guished.
Modelers also confront the issue of whether
sound-to-spelling conversion is serial (one pho-
neme at a time) or parallel. Our results suggest
that sound-to-spelling translation is not strictly a
phoneme-by-phoneme process. The choice of
graphemes for a given vowel is influenced by the
following context, implying that spellers do not
complete the selection of a spelling for a medial
vowel before they have processed the next pho-
neme. Most likely, our results suggest, other
phonemes from the same syllable are considered
when determining how to spell a particular pho-
neme. This conclusion is of particular interest
464 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468
given that one of the major models of the non-
lexical route in reading assumes serial, letter-by-
letter processing (Coltheart, Rastle, Perry, Lang-
don, & Ziegler, 2001).
Models of the spelling process must also con-
front the issue of whether feedback from reading
plays a role in spelling and, if so, how and when.
This issue is analogous to the one of whether
feedback from spelling plays a role in reading, an
issue that has led to some controversy (Peereman,
Content, & Bonin, 1998; Ziegler, Stone, & Jacobs,
1997). As discussed above, our results suggest
that the end result in spelling production can be
influenced by both knowledge of sound-to-spell-
ing translation (a feedforward effect) and knowl-
edge of common letter patterns (a feedback
effect).
Our results have implications for the study of
spelling development as well as for the modeling
of skilled spelling. If experienced spellers attempt
to cope with the challenges posed by the English
writing system by using context that extends be-
yond the single phoneme, then what about less
experienced spellers? One hypothesis is that
spelling development, like reading development,
proceeds from small units to larger units (e.g.,
Frith, 1985). Our results support the idea that
larger units are important in adult spelling pro-
duction, contrary to some previous claims (Barry
& Seymour, 1988; Perry et al., 2002). If these units
become important only for experienced spellers,
then young children should not show contextual
effects of the kinds documented here. Another
hypothesis is that large units, especially rimes, are
used from the beginning of literacy development
(e.g., Goswami, 1988). On this view, contextual
effects should be found in young children. We are
currently carrying out research to address these
questions.
Our results further suggest that care needs to
be taken in classifying words for studies of spell-
ing and reading. The assumption that spellers
transcribe phonemes without regard to context
may have led to problems in some previous work.
For example, Kreiner and Gough (1990) used a
measure of the relative probability of phoneme–
grapheme correspondences that assumed that
spellers consider the spelling options for a pho-
neme in proportion to their frequency in the lex-
icon as a whole. If spellers give most weight to the
options that are appropriate for a particular
consonantal context, however, such a measure
may not be ideal. As another example, Waters,
Bruck, and Malus-Abramowitz (1988) included a
category of words (labeled regular* words) in
their study of third- to fifth-grade children in
which the words had one or more problem pho-
nemes that could be represented by several dif-
ferent graphemes. The researchers expected
children to perform poorly on these words, as
there appeared to be no principled basis for se-
lection of the correct grapheme from among the
several possibilities. However, the children per-
formed better than expected on the regular*
words. The children may have been able to use
statistical patterns of the kind examined here to
boost their performance on these words. For ex-
ample, food was considered to be a regular* word
because /u/ has several possible spellings. How-
ever, oo is by far the most common spelling of /u/
after noncoronal consonant such as /f/, and chil-
dren may not have given serious consideration to
other options.
In other cases, problems may arise because
of an assumption that spellers use rime-based
units but not other large units (Stone, Vanhoy,
& Van Orden, 1997; Ziegler et al., 1997). In
their study of feedback effects in word reading,
for example, Ziegler et al. classified = Ap= as an
inconsistently spelled rime because it may be
spelled as ap (as in swap) as well as op (as in
chop). However, the a spelling only occurs after
w and so the rime may not be inconsistent for
those who are sensitive to this pattern. Measures
of sound–spelling consistency that do not con-
sider onset-to-vowel associations may not pro-
vide an ideal basis for selecting stimuli for
experiments or drawing conclusions about un-
derlying processes.
Our findings also have implications for what
makes some people good spellers and others poor
spellers. Traditionally, differences in spelling
ability have been attributed to differences in vi-
sual memory ability (see Fischer et al., 1985). The
assumption has been that the English writing
system is so irregular and so illogical that the only
way to become a skilled speller is through rote
memorization. However, research has not shown
a clear connection between nonverbal memory
skills and spelling ability (Fischer et al., 1985;
Giles & Terrell, 1997). The results of Experiment
3 provide initial evidence that good spellers at the
college level are more sensitive than poor spellers
to the contextual factors influencing vowel rep-
resentation. That is, good spellers have a higher
degree of linguistic sensitivity than poor spellers
do. The results of Fischer et al. (1985) point to
a similar conclusion. Our results suggest that
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 465
differences in linguistic sensitivity are found for
monomorphemic words, not just for the mor-
phologically complex words that were the focus
of the Fischer et al. research. Moreover, our re-
sults point to some of the relevant linguistic di-
mensions.
In our view, research and modeling in the area
of spelling (and in reading) must be based on a
good understanding of the characteristics of the
writing system under study. The finding that the
spelling of English vowels can be systematically
affected by the identity of the surrounding con-
sonants (Kessler & Treiman, 2001) set the stage
for the present demonstration that adult spellers
are attuned to context in their spelling of vowels.
This combination of statistical studies of language
and behavioral work holds promise for studies of
spelling and reading, as for psycholinguistic re-
search more generally. We look forward to further
applications of this approach in the study of more
complex words and in the study of spelling de-
velopment.
Appendix A. Stimuli for Experiment 1
Case 1: eExperimental: kled; g red; st red; ged; gled; jed;hed; ved; kwed; smedControl: kleb; g redZ; st reg; getS; glep; jeb; hek;vep; kweg; smek
Case 2: i
Experimental: S rid; fip; gid; jid; snip; p rid; hip;sk rip; smip; zip
Control: S rig; fih; gim; jið; snim; p rig; hist;sk rih; smið; zist
Case 3: o
Experimental: dwol; g rol; jol; klol; kwol; plol;
skwol; snol; sp rol; wol
Control: dwot; g
rok; jop; klob; kwon; plotS;skwof; snoð; sp roð;wog
Case 4: u
Experimental: SutS; fuS; juS; klus; lutS; hul; sul;
slutS; smul; zul
Control: Suf ; fuf; juf; kluf ; luf ; huk; suf; sluk;
smuf ; zuf
Case 5: au
Experimental: Saul; maul; naul; haun; vaun; zaul;spaul; p raun; b raul; smaun
Control: SaudZ; mautS; naudZ; haus; vaus; zautS;
spaudZ; p raudZ;b rautS; smautS
Case 6: ai
Experimental: d rait; glait; g rait; gait; jait; hait;
stait; zait; p rait; pait
Control: d raib; glaib; g raiv; gaif ; jais; haif; staib;
zaif ; p raib; paif
Fillers
his; f r
ip; des; fesk; gl�h; h rVd; h c
if; g
crb; flVn;
naið; S
c
in; splaut; jit; S rig; zeb; skent; v�sp;
g
r�k; d c; spV¢
Stimuli for Experiment 2
Case 1: A
Experimental: skw Anz; kw Ats; kw Abd; w Abd;
tw An; w AdZ; skw Amp; kw AtS; kw Ap; gw At
Control: sl Anz; k r Ats; kl Abd; t r Abz; gl An; bl AdZ;n Amp; fl AtS; bl Ap; k r At
Case 2: g
Experimental: wgdZ; wgp; wgtS; wgg; wgS;wgf; wgb; wgn; wgpt; wghtControl: kgdZ; kgp; kgtS; kgg; kgS; blgf ; slgb;
blgn; kgpt; kghtCase 3: u
Experimental: budZ; guh; muk; hus; hun; sput;skudZ; smud; swu; muhControl: fludZ; S ruh; p ruk; klus; plun; S rut; f rudZ;
flud; plu; ðuhFillers
bis; b r
ip; dek; feg; fl�h; gVd; hob; k crb; lVn;
maið; p c
in; splau; rit; Sig; seb; snep; t�sp; t rok;v c; zV¢
Stimuli for Experiment 3
Coda-to-vowel associations
Case 1, experimental: shred, bled, moped
Case 1, control: fleck, wretch, trek
Case 2, experimental: centipede, mead, plead, reap,
impede, stampede
Case 2, control: academe, ream, sheath, bequeath,
blaspheme, theme
Case 3, experimental: cajole, camisole, casse-
role, console, Creole, dole, oriole, parole, shoal,
tadpole
Case 3, control: abode, anecdote, isotope, cy-
clone, microbe, lobe, antidote, corrode, croak,
diode
Case 4, experimental: swoosh
Case 4, control: nook
Case 5, experimental: afoul, pronoun
Case 5, control: vouch, slouch
Case 6, experimental: contrite, extradite, finite,
ignite, parasite, smite, sprite, termite
Case 6, control: confide, pesticide, capsize, entice,
diatribe, fife, snide, chastise
Onset-to-vowel associations:
Case 1, experimental: wok, wombat, whopper,
wonk
Case 1, control: cog, possum, nozzle, pomp
Case 2, experimental: whir, whirlpool
Case 2, control: blurt, kerchief
466 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468
Case 3, experimental: Budapest, cuckoo, Fuji,
guru, scuba, Buddha
Case 3, control: jubilant, sushi, Juneau, judo,
sumo, frugal
Stimuli for Experiment 4
Case 2, /i/
Experimental, same syllable: =S rid0bok=; =snip0n c
i=;
=hip0neb=; =sk rip0t A
ud=; =smip0n A
us=; =zip0nef=;=klid0l c
i=; =ip0dek=Control, same syllable: =S rig0bok=; =snim0n c
i=;
=hig0neb=; =sk rih0t A
ud=; =smið0n A
us=; =zim0nef=;=klih0l c
i=; =ig0dek=Experimental, different syllable: =S ri0dok=; =sni0p c
i=;=hi0peb=; =sk ri0p A
ud=; =smi0p A
us=; =zi0pef=;=kli0d c
i=; =i0pek=Control, different syllable: =S ri0gok=; =sni0m c
i=;
=hi0geb=; =sk ri0h A
ugd=; =smi0ð A
us=; =zi0mef=;=kli0h c
i=; =i0gek=
Case 5, =au=
Experimental, same syllable: =Saul0wost=;
=zaul0wug=; =spaul0winh=; =smaun0leb=;=faun0lib=; =aun0lef=; =laun0luf=; =graun0l�sp=
Control, same syllable: =SaudZ0wost=; =zautS0wug=;=spaudZ0winh=; =smautS0leb=; =fautS0lib=;=audZ0lef=; =lautS0luf=; =graudZ0l�sp=
Experimental, different syllable: =Sau0lost=;=zau0lug=; =spau0linh=; =smau0neb=; =fau0nib=;=au0nef=; =lau0nuf=; =grau0n�sp=
Control, different syllable: =Sau0dZost=;=zau0tSug=; =spau0dZinh=; =smau0tSeb=; =fau0tSib=;=au0dZef=; =lau0tSuf=; =grau0dZ�sp=
Case 6, =ai=
Experimental, same syllable: =glait0n A
us=;
=g rait0k A
S=; =zait0p A
ugd=; =st rait0lef=; =S rait0mis=;
=jait0fug=; =klait0n An=; =dZait0keb=Control, same syllable: =glaib0n A
us=;=g raiv0k A
S=; =zaif 0p A
ugd=; =st raiz0lef=;=S raif 0mis=; =jaib0fug=; =klaib0n An=; =dZaif 0keb=Experimental, different syllable: =glai0t A
us=;=g rai0t A
S=; =zai0t A
ud=; =st rai0tef=; =S rai0tis=;=jai0tug=; =klai0t An=; =dZai0teb=Control, different syllable: =glai0b A
us=; =g rai0v A
S=;
=zai0f A
ud=; =st rai0zef=; =S rai0fis=; =jai0bug=;=klai0b An=; =dZai0feb=
Fillers
=0bintE=; =0t r�b=; =0tipko=; =0puk=; =0fegis=;
=0g�nid=; =0pl AksE=; =0flVn=; =0wikEl=; =0p�sk=;
=0gimbE=; =0lVh=; =0mul=; =0dil�b=; =0ikin=;
=0est�k=; =0p�do=; =0S rV¢=; =0tivEt=; =0k r Ab=
References
Andrews, S., & Scarratt, D. R. (1998). Rule and analogy
mechanisms in reading nonwords: Hough dou peapel
rede gnew wirds? Journal of Experimental Psychol-
ogy: Human Perception and Performance, 24, 1052–
1086.
Barry, C., & Seymour, P. H. K. (1988). Lexical priming
and sound-to-spelling contingency effects in nonword
spelling. Quarterly Journal of Experimental Psychol-
ogy, 40A, 5–40.
Campbell, R. (1983). Writing nonwords to dictation.
Brain and Language, 19, 153–178.
Coltheart, M., Rastle, K., Perry, C., Langdon, R., &
Ziegler, J. (2001). DRC: a dual route cascaded model
of visual word recognition and reading aloud.
Psychological Review, 108, 204–258.
Fischer, F. W., Shankweiler, D., & Liberman, I. Y.
(1985). Spelling proficiency and sensitivity to word
structure. Journal of Memory and Language, 24, 423–
441.
Frith, U. (1985). Beneath the surface of developmental
dyslexia. In K. E. Patterson, J. C. Marshall, & M.
Coltheart (Eds.), Surface dyslexia: Neuropsycholog-
ical and cognitive studies of phonological reading (pp.
301–330). London: Erlbaum.
Fudge, E. C. (1969). Syllables. Journal of Linguistics, 5,
253–286.
Giles, D. C., & Terrell, C. D. (1997). Visual sequential
memory and spelling ability. Educational Psychology,
17, 245–253.
Glushko, R. J. (1979). The organization and activation
of orthographic knowledge in reading aloud. Journal
of Experimental Psychology: Human Perception and
Performance, 5, 674–691.
Goodman, R. A., & Caramazza, A. (1986). Aspects of
the spelling process: evidence from a case of acquired
dysgraphia. Language and Cognitive Processes, 1,
263–296.
Goswami, U. (1988). Children�s use of analogy in
learning to spell. British Journal of Developmental
Psychology, 6, 21–33.
Kessler, B., & Treiman, R. (1997). Syllable structure and
the distribution of phonemes in English syllables.
Journal of Memory and Language, 37, 295–311.
Kessler, B., & Treiman, R. (2001). Relationships be-
tween sounds and letters in English monosyllables.
Journal of Memory and Language, 44, 592–617.
Kreiner, D. S., & Gough, P. B. (1990). Two ideas about
spelling: rules and word-specific memory. Journal of
Memory and Language, 29, 103–118.
Martensen, H., Maris, E., & Dijkstra, T. (2000). When
does inconsistency hurt? On the relation between
phonological consistency effects and the reliability
of sublexical units. Memory & Cognition, 28, 648–
656.
Mewhort, D. J. K., & Beal, A. L. (1977). Mechanisms of
word identification. Journal of Experimental Psy-
chology: Human Perception and Performance, 3, 629–
640.
Olson, A., & Caramazza, A. (1994). Representation and
connectionist models: the NETspell experience. In G.
D. A. Brown & N. C. Ellis (Eds.), Handbook of
R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468 467
spelling: Theory, process and intervention (pp. 337–
363). Chichester, UK: Wiley.
Peereman, R., Content, A., & Bonin, P. (1998). Is
perception a two-way street? The case of feedback
consistency in visual word recognition. Journal of
Memory and Language, 39, 151–174.
Perry, C., Ziegler, J. C., & Coltheart, M. (2002).
Dissociation between orthographic awareness and
spelling production. Applied Psycholinguistics, 23,
43–73.
Prinzmetal, W., Treiman, R., & Rho, S. (1986). How to
see a reading unit. Journal of Memory and Language,
25, 461–475.
Sanders, R. J., & Caramazza, A. (1990). Operation of
the phoneme-to-grapheme conversion mechanism in
a brain injured patient. Reading and Writing: An
Interdisciplinary Journal, 2, 61–82.
Selkirk, E. O. (1982). The syllable. In H. Van der Hulst
& N. Smith (Eds.), The structure of phonological
representations (Part II, pp. 337–383). Dordrecht,
The Netherlands: Foris.
Stone, G. O., Vanhoy, M., & Van Orden, G. C. (1997).
Perception is a two-way street: feedforward and
feedback phonology in visual word recognition.
Journal of Memory and Language, 36, 337–359.
Taraban, R., & McClelland, J. L. (1987). Conspiracy
effects in word pronunciation. Journal of Memory
and Language, 26, 608–631.
Treiman, R. (1983). The structure of spoken syllables:
evidence from novel word games. Cognition, 15, 49–
74.
Treiman, R. (1993). Beginning to spell: A study of first-
grade children. New York: Oxford University Press.
Treiman, R., & Danis, C. (1988a). Short-term memory
errors for spoken syllables are affected by the
linguistic structure of the syllables. Journal of
Experimental Psychology: Learning, Memory, and
Cognition, 14, 145–152.
Treiman, R., & Danis, C. (1988b). Syllabification of
intervocalic consonants. Journal of Memory and
Language, 27, 87–104.
Treiman, R., Mullennix, J., Bijeljac-Babic, R., & Rich-
mond-Welty, E. D. (1995). The special role of rimes
in the description, use, and acquisition of English
orthography. Journal of Experimental Psychology:
General, 124, 107–136.
Treiman, R., & Zukowski, A. (1988). Units in reading
and spelling. Journal of Memory and Language, 27,
466–477.
Treiman, R., & Zukowski, A. (1990). Toward an
understanding of English syllabification. Journal of
Memory and Language, 29, 66–85.
Ward, G. (1993). Moby Pronunciator II. [Data file].
Available: ftp://svr-ftp.eng.cam.ac.uk/pub/comp.-
speech/dictionaries/moby/.
Waters, G. S., Bruck, M., & Malus-Abramowitz, M.
(1988). The role of linguistic and visual information
in spelling: a developmental study. Journal of
Experimental Child Psychology, 45, 400–421.
Weide, R. L. (Ed.). (1995). The Carnegie Mellon
pronouncing dictionary: cmudict.0.4 [Data file].
Available: ftp://svr-ftp.eng.cam.ac.uk/pub/comp.-
speech/dictionaries/cmudict/.
Ziegler, J. C., Stone, G. O., & Jacobs, A. M. (1997).
What is the pronunciation for -ough and the spelling
for /u/? A database for computing feedforward and
feedback consistency in English. Behavior Research
Methods, Instruments, & Computers, 29, 600–618.
Zeno, S. M., Ivenz, S. H., Millard, R. T., & Duvvuri, R.
(1995). Educator�s word frequency guide. Brewster,
NY: Touchstone Applied Science Associates.
Zorzi, M., Houghton, G., & Butterworth, B. (1998).
Two routes or one in reading aloud? A connectionist
dual-process model. Journal of Experimental Psy-
chology: Human Perception and Performance, 24,
1131–1161.
468 R. Treiman et al. / Journal of Memory and Language 47 (2002) 448–468